The recent surge in cyberattacks across Japan, combined with evidence of AI-assisted attacks against financial institutions in South Korea, raises an uncomfortable question about how organizations evaluate cybersecurity risk. We have spent decades developing security frameworks, establishing control requirements and building processes intended to manage exposure. Much of that work assumes organizations have a reasonable opportunity to identify weaknesses and address them before attackers can exploit them. Artificial intelligence is beginning to challenge those assumptions by changing the speed and economics of cyberattacks, potentially making established security practices less effective even when organizations continue following them.

What concerns me most is the possibility that businesses will continue measuring cybersecurity effectiveness against requirements that no longer reflect the conditions in which they operate. A control can remain compliant with an established standard while the protection it provides becomes less reliable. Management may receive favorable assessment results without recognizing that the underlying assumptions about attacker capabilities have changed. This creates a growing gap between what an organization believes its cybersecurity program can accomplish and what it may actually be capable of defending against.

The developments in Asia provide a useful illustration of that problem because they show both the growing scale of cyber incidents and the emerging role of AI in making attacks more accessible. They also demonstrate why we need to be careful about distinguishing confirmed evidence from assumptions about AI involvement.

Japan's Cybersecurity Crisis

On October 9, 2026, Japan's ruling Liberal Democratic Party convened an emergency meeting of its National Cybersecurity Strategy Headquarters following a series of attacks against major Japanese businesses and public institutions. According to reporting by the Financial Times, more than 20 major companies had reported cybersecurity incidents in recent weeks, affecting organizations across retail, transportation, financial services and other industries. The incidents raised concerns about the exposure of substantial volumes of customer information and prompted senior Japanese officials to describe the situation as a cybersecurity emergency.

The affected organizations included car-sharing provider Times Car, which reported potential exposure involving approximately 6.6 million customer accounts, and railway operator JR East, which disclosed that information associated with approximately 1.67 million customer email addresses may have been compromised. The incidents also raised concerns about the consequences of attacks against technology providers whose services support multiple organizations. Some reported data exposures were connected to a ransomware incident involving IDC Frontier, a cloud services subsidiary of SoftBank. The breadth of the incidents illustrates how cybersecurity exposure can extend beyond an organization's directly managed systems into the services and infrastructure on which its operations depend.

On October 12, Trend Micro published additional research showing that Japan had recorded more publicly disclosed cybersecurity incidents during the first nine months of 2026 than during all of 2025. September alone accounted for 86 incidents, representing an increase of approximately 18% from August and 37% from July. Its incident response researchers also identified attackers exploiting known vulnerabilities in publicly accessible applications to establish access to corporate environments. The findings suggest that many organizations continue struggling with weaknesses that cybersecurity programs have understood for years, even as the number of potential attackers and the speed at which they can operate continue to increase.

The company's October 12 threat advisory was careful to distinguish the broader threat posed by AI from the specific incidents occurring in Japan. Researchers had not established that AI was involved in the recent Japanese attacks. They did, however, warn that AI can make reconnaissance and vulnerability analysis more efficient, reducing the resources attackers need to identify weaknesses across large numbers of organizations. That distinction matters because the concern is supported by the changing capabilities available to attackers without requiring us to attribute every new incident to artificial intelligence.

The Attack Surface Is Becoming Easier to Understand

On October 13, Japan's National Institute of Information and Communications Technology published research identifying a technique associated with some of the country's recent data breaches. Its researchers described what they call an App-to-Backend attack, or A2B, in which attackers analyze publicly available Android applications to identify information about the servers and application programming interfaces supporting those applications. That information can then be used to develop attacks against backend systems that organizations may not realize are exposing useful technical details through their public applications.

The NICT research announcement explains that attackers can examine how an application communicates with its supporting infrastructure and use that knowledge to identify weaknesses in systems that were never intended to be publicly accessible. The researchers warned that similar exposure could exist in other mobile applications and web services. They also identified AI as a technology capable of making application analysis and vulnerability discovery more efficient, potentially allowing these techniques to be applied against a much larger population of targets. The research does not establish that AI was used in every identified A2B incident, but it provides a concrete example of an attack process that AI could accelerate.

This is an important development because many organizations evaluate their attack surface primarily through the systems and services they know are exposed to the internet. They may conduct vulnerability scans and penetration tests against identified assets while overlooking information that can be obtained by analyzing the applications those systems support. The organization may believe it has a reasonable understanding of its external exposure, but attackers can develop a different picture by combining information from sources that are individually available to the public.

The underlying weakness is not necessarily new. What changes with AI is the amount of effort required to discover and exploit it. Techniques that once demanded considerable technical expertise and manual investigation can become more accessible when portions of that work are automated. This creates an environment in which the ability to find vulnerabilities may improve faster than an organization's ability to recognize and remediate them.

South Korea Shows How AI Can Change Attacker Capabilities

The evidence of AI-assisted attacks is more direct in neighboring South Korea. On October 8, Reuters reported that cybersecurity researchers at CrowdStrike had linked attacks against South Korean banks to a suspected individual using AI-assisted penetration-testing tools. The investigation identified the use of ARTEX, an AI-enabled security testing agent, alongside Anthropic's Claude Code. CrowdStrike assessed that the suspected attacker was likely operating from China and using these tools to conduct attacks against multiple financial institutions.

At least nine South Korean banks had been identified as targets when the findings were reported. Shinhan Bank disclosed that personal information associated with approximately 25,000 customers had been compromised, while other institutions also reported exposures. CrowdStrike described the activity as an example of a human attacker using AI agents to expand the number of organizations that could be targeted within a relatively short period. The distinction is important because the evidence points to AI-assisted activity directed by a person, rather than establishing that an autonomous system independently selected and attacked the banks.

The situation continued developing. On October 14, Reuters reported that South Korea's Financial Services Commission had identified hacking incidents involving 16 financial firms and approximately 100,000 cases of leaked information. Regulators announced additional investigations into the methods used, including whether AI had been involved. The findings should not be interpreted as confirmation that every incident was connected to the same attacker or that all of the activity was AI-assisted. They do, however, demonstrate the scale of the problem confronting financial institutions and the increasing importance regulators are placing on understanding the methods behind these attacks.

What makes the South Korean case particularly relevant is the apparent ability of an individual attacker to use AI tools to perform work that would previously have required more specialized expertise. The attacker still needed access to appropriate tools and some understanding of how to use them, but AI appears to have expanded the scope of activity that could be conducted. If that capability becomes more widely available, organizations may face more frequent attempts to exploit weaknesses that have historically received less attention because attackers lacked the resources to pursue them.

When Established Security Processes Become Outdated

Most organizations establish cybersecurity processes around assumptions about the threat environment. Vulnerability management programs define how frequently systems should be assessed and how quickly identified weaknesses should be remediated. Security teams establish monitoring practices based on the systems they consider most exposed. Management approves budgets and staffing levels intended to support those activities, often using established frameworks and industry benchmarks to determine whether the resulting program is reasonable.

Those decisions become part of the organization's operating model and may remain largely unchanged for years. Periodic assessments can confirm that the required activities are occurring, but they don't necessarily establish that the processes remain appropriate for the conditions the organization currently faces. A company may successfully complete quarterly vulnerability assessments and meet its documented remediation targets while remaining exposed to weaknesses that attackers can now identify and exploit much more quickly.

Consider an organization that allows 30 days to remediate certain vulnerabilities after discovery. That timeline may reflect a reasonable balance between security requirements and the operational effort required to test and deploy changes. However, if attackers can discover exposed systems and identify exploitable weaknesses in a fraction of the time previously required, the organization's approved remediation process may leave it vulnerable for much of that period. The process continues functioning as designed, but the level of protection management expects from it may no longer be justified.

This is the problem with relying too heavily on compliance as evidence of security effectiveness. Compliance demonstrates that an organization is meeting an established requirement. Whether that requirement continues to provide adequate protection depends on the conditions under which the control operates. AI is beginning to change some of those conditions, and organizations need a way to recognize when their established practices are no longer producing the intended result.

The Decision Debt Behind Cybersecurity Exposure

I have written extensively about decision debt because organizations often accumulate risk through decisions that were reasonable when they were made but have not been revisited as circumstances change. Cybersecurity provides numerous examples because control requirements are frequently embedded in policies and operating procedures that evolve more slowly than the technologies and threats they are intended to address.

An organization may establish a vulnerability management process based on its available resources and the level of exposure it believes it faces. Over time, the business introduces new applications and becomes more dependent on external services. The attack surface expands, but the original remediation timelines and assessment methods remain in place because they continue satisfying the organization's documented requirements. Management receives reports showing that the cybersecurity program is operating according to policy without necessarily recognizing that the policy itself may no longer reflect the business's exposure.

AI can accelerate the consequences of these deferred decisions. When attackers become more capable of discovering weaknesses and conducting reconnaissance at scale, the time available for organizations to recognize and address exposure may shrink. Decisions about assessment frequency and remediation priorities become more consequential, particularly when the business has developed dependencies that were not fully understood when those decisions were originally made.

The debt becomes visible when an incident occurs and management discovers that the organization was following its approved processes while remaining exposed to a risk those processes were supposed to manage. Addressing that problem requires more than updating a security policy after an incident. It requires establishing a mechanism for continuously evaluating whether the assumptions behind existing decisions remain valid.

Integrated Assurance Must Account for a Changing Threat Environment

This is where I believe Integrated Assurance becomes increasingly important. Organizations have invested heavily in systems for monitoring cybersecurity controls and measuring compliance, but they often lack a reliable way to connect those measurements to the operating conditions that determine whether the controls remain effective.

Integrated Assurance is intended to provide that connection by bringing together visibility into the organization's environment, validation of its controls and an understanding of the business outcomes those controls are expected to support. The objective is to establish whether management can continue relying on its cybersecurity capabilities as the organization and its exposure change.

In the context of AI-assisted attacks, that means evaluating whether the processes used to identify and remediate vulnerabilities remain responsive to the current threat environment. It also means understanding where the organization has become dependent on applications and external services whose exposure may not be adequately represented in traditional security assessments.

The Japanese incidents illustrate why this broader perspective matters. An organization may have a well-managed internal security environment while remaining vulnerable through an application or technology provider that supports an important business capability. The security team may have limited visibility into the provider's operating environment, while the business has become increasingly dependent on the services being delivered.

Assurance needs to account for those relationships because the consequences of a cyber incident are determined by how the business operates. A vulnerability affecting a relatively small supplier may become material if that supplier supports a capability the organization cannot readily replace. AI-assisted reconnaissance could make those suppliers more attractive targets by reducing the cost of identifying and exploiting their weaknesses.

The challenge for management is understanding whether the organization's existing controls and recovery capabilities remain appropriate for the dependencies it has accumulated. That requires a more complete view of the operating environment than most traditional cybersecurity assessments provide.

Cyber Insurance Is Facing the Same Problem

The implications for cyber insurance are significant because underwriting depends on understanding the relationship between an organization's controls and the likelihood and severity of a loss. Insurers routinely ask applicants about multifactor authentication, endpoint protection, vulnerability management, backups and incident response capabilities because those controls have demonstrated value in managing cyber exposure.

The challenge is that the effectiveness of a control depends partly on the environment in which it operates. A vulnerability management process that was appropriate several years ago may provide less protection if attackers can identify exploitable weaknesses more quickly. A company may accurately represent that it performs regular security assessments without recognizing that the frequency or scope of those assessments no longer reflects its exposure.

This creates a potential problem for both the insurer and the insured. The insurer may rely on representations that accurately describe the organization's security activities but provide an incomplete picture of how effectively those activities manage current risk. The business may believe it has satisfied the insurer's requirements without understanding whether its controls remain sufficient to protect the operations on which it depends.

The emergence of AI-assisted attacks is also beginning to attract regulatory attention. On October 13, Reuters reported that banking regulators in the United States and Europe were increasing their scrutiny of advanced AI systems and the risks they could introduce into financial institutions. The concerns included the ability to control autonomous systems and the adequacy of recovery plans if AI-related incidents create widespread disruption. Regulators were also examining dependencies on external technology providers, recognizing that failures involving shared infrastructure could affect multiple institutions.

These developments suggest that cyber insurance readiness will increasingly need to involve evidence about how controls perform rather than relying exclusively on whether they have been implemented. There is not yet sufficient evidence to conclude that the Japanese or South Korean incidents will produce specific changes in underwriting requirements, but the broader direction is consistent with insurers placing greater emphasis on continuous visibility and control effectiveness.

This is one of the arguments behind Can We Insure This? A cyber insurance application represents management's understanding of how the organization manages risk. The quality of those representations depends on whether the business has enough visibility and evidence to support them. As the threat environment changes, organizations will need to become more disciplined about validating that the controls described in their applications continue providing the protection those answers imply.

Business Survivability Changes the Question

There is a broader issue here that extends beyond the effectiveness of individual cybersecurity controls. As AI increases the speed and accessibility of cyberattacks, organizations may have less time to recognize and contain incidents before they begin affecting operations. That makes the ability to continue operating during a disruption increasingly important.

A business can have a cybersecurity program that meets established requirements and still experience a serious interruption because an attacker exploits a weakness in a critical application or technology provider. The financial consequences may depend less on the sophistication of the attack than on how quickly the organization can restore the capabilities required to serve customers and meet its obligations.

Business Survivability requires understanding those capabilities and the dependencies supporting them before an incident occurs. Management needs to know which operations are essential and whether the organization has realistic alternatives when the systems supporting them become unavailable. Recovery planning must account for the possibility that a critical provider is experiencing the same disruption or that the organization cannot immediately restore the technology it normally depends upon.

AI adds another dimension because it can affect both sides of that equation. Attackers may use AI to identify and exploit weaknesses more efficiently, while businesses are introducing AI into their own operations and becoming dependent on services whose reliability and security they may not fully understand. The resulting exposure can grow without a corresponding improvement in management's understanding of what the business needs to survive a disruption.

Insurance remains important because it can help finance recovery from covered events, but coverage does not establish that the organization can continue operating while recovery takes place. A business that has satisfied its insurer's underwriting requirements may still face an interruption that exceeds its operational or financial tolerance.

That is why I believe the conversation needs to move beyond whether organizations have implemented the expected cybersecurity controls. Management needs reliable evidence that those controls, together with the organization's operating and recovery capabilities, remain sufficient for the business conditions it faces.

The Risk of Continuing to Measure Yesterday's Decisions

The developments in Japan and South Korea should concern business leaders because they illustrate how quickly the environment surrounding established cybersecurity practices can change. Japan's recent incidents demonstrate the scale of exposure that can exist across organizations and their technology dependencies. The South Korean banking attacks provide more direct evidence that AI-assisted tools can expand the capabilities of individual attackers. Together, they raise questions about whether cybersecurity programs are adapting quickly enough to remain effective.

I don't believe the answer is to abandon established security frameworks or assume that every organization needs an entirely new cybersecurity strategy. Many of the weaknesses being exploited remain familiar, and the controls required to address them are well understood. The problem is that organizations may continue operating those controls according to assumptions that no longer reflect how quickly attackers can act or how dependent the business has become on the systems being protected.

The more important work involves examining whether existing processes remain appropriate and identifying where decisions need to be reconsidered. An organization that has historically accepted a particular remediation timeline should understand whether that timeline continues to provide reasonable protection. A business that relies on external technology providers should understand whether those dependencies introduce exposure that its cybersecurity program is capable of detecting and managing.

This is ultimately a governance issue because the decisions that determine cybersecurity effectiveness extend beyond the security team. Management establishes the organization's tolerance for risk and approves the investments required to manage it. Those decisions need to be informed by evidence about current operating conditions rather than relying on the continued execution of processes that may have been established years earlier.

AI is making that responsibility more urgent because it can accelerate the consequences of weaknesses that organizations have already accepted or failed to recognize. A company may remain compliant with its policies and still discover that the protection those policies were intended to provide has deteriorated.

For me, that is the most important lesson emerging from these incidents. The question is whether management has enough assurance to know when its existing cybersecurity practices are no longer producing the outcomes the business depends upon. As AI changes the capabilities available to attackers, organizations need to become equally disciplined about challenging the assumptions behind their own decisions.